Stay ahead of learning milestones! Enroll in a class over the summer!

G
Topic
First Poster
Last Poster
k a April Highlights and 2025 AoPS Online Class Information
jlacosta   0
Apr 2, 2025
Spring is in full swing and summer is right around the corner, what are your plans? At AoPS Online our schedule has new classes starting now through July, so be sure to keep your skills sharp and be prepared for the Fall school year! Check out the schedule of upcoming classes below.

WOOT early bird pricing is in effect, don’t miss out! If you took MathWOOT Level 2 last year, no worries, it is all new problems this year! Our Worldwide Online Olympiad Training program is for high school level competitors. AoPS designed these courses to help our top students get the deep focus they need to succeed in their specific competition goals. Check out the details at this link for all our WOOT programs in math, computer science, chemistry, and physics.

Looking for summer camps in math and language arts? Be sure to check out the video-based summer camps offered at the Virtual Campus that are 2- to 4-weeks in duration. There are middle and high school competition math camps as well as Math Beasts camps that review key topics coupled with fun explorations covering areas such as graph theory (Math Beasts Camp 6), cryptography (Math Beasts Camp 7-8), and topology (Math Beasts Camp 8-9)!

Be sure to mark your calendars for the following events:
[list][*]April 3rd (Webinar), 4pm PT/7:00pm ET, Learning with AoPS: Perspectives from a Parent, Math Camp Instructor, and University Professor
[*]April 8th (Math Jam), 4:30pm PT/7:30pm ET, 2025 MATHCOUNTS State Discussion
April 9th (Webinar), 4:00pm PT/7:00pm ET, Learn about Video-based Summer Camps at the Virtual Campus
[*]April 10th (Math Jam), 4:30pm PT/7:30pm ET, 2025 MathILy and MathILy-Er Math Jam: Multibackwards Numbers
[*]April 22nd (Webinar), 4:00pm PT/7:00pm ET, Competitive Programming at AoPS (USACO).[/list]
Our full course list for upcoming classes is below:
All classes run 7:30pm-8:45pm ET/4:30pm - 5:45pm PT unless otherwise noted.

Introductory: Grades 5-10

Prealgebra 1 Self-Paced

Prealgebra 1
Sunday, Apr 13 - Aug 10
Tuesday, May 13 - Aug 26
Thursday, May 29 - Sep 11
Sunday, Jun 15 - Oct 12
Monday, Jun 30 - Oct 20
Wednesday, Jul 16 - Oct 29

Prealgebra 2 Self-Paced

Prealgebra 2
Sunday, Apr 13 - Aug 10
Wednesday, May 7 - Aug 20
Monday, Jun 2 - Sep 22
Sunday, Jun 29 - Oct 26
Friday, Jul 25 - Nov 21

Introduction to Algebra A Self-Paced

Introduction to Algebra A
Monday, Apr 7 - Jul 28
Sunday, May 11 - Sep 14 (1:00 - 2:30 pm ET/10:00 - 11:30 am PT)
Wednesday, May 14 - Aug 27
Friday, May 30 - Sep 26
Monday, Jun 2 - Sep 22
Sunday, Jun 15 - Oct 12
Thursday, Jun 26 - Oct 9
Tuesday, Jul 15 - Oct 28

Introduction to Counting & Probability Self-Paced

Introduction to Counting & Probability
Wednesday, Apr 16 - Jul 2
Thursday, May 15 - Jul 31
Sunday, Jun 1 - Aug 24
Thursday, Jun 12 - Aug 28
Wednesday, Jul 9 - Sep 24
Sunday, Jul 27 - Oct 19

Introduction to Number Theory
Thursday, Apr 17 - Jul 3
Friday, May 9 - Aug 1
Wednesday, May 21 - Aug 6
Monday, Jun 9 - Aug 25
Sunday, Jun 15 - Sep 14
Tuesday, Jul 15 - Sep 30

Introduction to Algebra B Self-Paced

Introduction to Algebra B
Wednesday, Apr 16 - Jul 30
Tuesday, May 6 - Aug 19
Wednesday, Jun 4 - Sep 17
Sunday, Jun 22 - Oct 19
Friday, Jul 18 - Nov 14

Introduction to Geometry
Wednesday, Apr 23 - Oct 1
Sunday, May 11 - Nov 9
Tuesday, May 20 - Oct 28
Monday, Jun 16 - Dec 8
Friday, Jun 20 - Jan 9
Sunday, Jun 29 - Jan 11
Monday, Jul 14 - Jan 19

Intermediate: Grades 8-12

Intermediate Algebra
Monday, Apr 21 - Oct 13
Sunday, Jun 1 - Nov 23
Tuesday, Jun 10 - Nov 18
Wednesday, Jun 25 - Dec 10
Sunday, Jul 13 - Jan 18
Thursday, Jul 24 - Jan 22

Intermediate Counting & Probability
Wednesday, May 21 - Sep 17
Sunday, Jun 22 - Nov 2

Intermediate Number Theory
Friday, Apr 11 - Jun 27
Sunday, Jun 1 - Aug 24
Wednesday, Jun 18 - Sep 3

Precalculus
Wednesday, Apr 9 - Sep 3
Friday, May 16 - Oct 24
Sunday, Jun 1 - Nov 9
Monday, Jun 30 - Dec 8

Advanced: Grades 9-12

Olympiad Geometry
Tuesday, Jun 10 - Aug 26

Calculus
Tuesday, May 27 - Nov 11
Wednesday, Jun 25 - Dec 17

Group Theory
Thursday, Jun 12 - Sep 11

Contest Preparation: Grades 6-12

MATHCOUNTS/AMC 8 Basics
Wednesday, Apr 16 - Jul 2
Friday, May 23 - Aug 15
Monday, Jun 2 - Aug 18
Thursday, Jun 12 - Aug 28
Sunday, Jun 22 - Sep 21
Tues & Thurs, Jul 8 - Aug 14 (meets twice a week!)

MATHCOUNTS/AMC 8 Advanced
Friday, Apr 11 - Jun 27
Sunday, May 11 - Aug 10
Tuesday, May 27 - Aug 12
Wednesday, Jun 11 - Aug 27
Sunday, Jun 22 - Sep 21
Tues & Thurs, Jul 8 - Aug 14 (meets twice a week!)

AMC 10 Problem Series
Friday, May 9 - Aug 1
Sunday, Jun 1 - Aug 24
Thursday, Jun 12 - Aug 28
Tuesday, Jun 17 - Sep 2
Sunday, Jun 22 - Sep 21 (1:00 - 2:30 pm ET/10:00 - 11:30 am PT)
Monday, Jun 23 - Sep 15
Tues & Thurs, Jul 8 - Aug 14 (meets twice a week!)

AMC 10 Final Fives
Sunday, May 11 - Jun 8
Tuesday, May 27 - Jun 17
Monday, Jun 30 - Jul 21

AMC 12 Problem Series
Tuesday, May 27 - Aug 12
Thursday, Jun 12 - Aug 28
Sunday, Jun 22 - Sep 21
Wednesday, Aug 6 - Oct 22

AMC 12 Final Fives
Sunday, May 18 - Jun 15

F=ma Problem Series
Wednesday, Jun 11 - Aug 27

WOOT Programs
Visit the pages linked for full schedule details for each of these programs!


MathWOOT Level 1
MathWOOT Level 2
ChemWOOT
CodeWOOT
PhysicsWOOT

Programming

Introduction to Programming with Python
Thursday, May 22 - Aug 7
Sunday, Jun 15 - Sep 14 (1:00 - 2:30 pm ET/10:00 - 11:30 am PT)
Tuesday, Jun 17 - Sep 2
Monday, Jun 30 - Sep 22

Intermediate Programming with Python
Sunday, Jun 1 - Aug 24
Monday, Jun 30 - Sep 22

USACO Bronze Problem Series
Tuesday, May 13 - Jul 29
Sunday, Jun 22 - Sep 1

Physics

Introduction to Physics
Wednesday, May 21 - Aug 6
Sunday, Jun 15 - Sep 14
Monday, Jun 23 - Sep 15

Physics 1: Mechanics
Thursday, May 22 - Oct 30
Monday, Jun 23 - Dec 15

Relativity
Sat & Sun, Apr 26 - Apr 27 (4:00 - 7:00 pm ET/1:00 - 4:00pm PT)
Mon, Tue, Wed & Thurs, Jun 23 - Jun 26 (meets every day of the week!)
0 replies
jlacosta
Apr 2, 2025
0 replies
k i Adding contests to the Contest Collections
dcouchman   1
N Apr 5, 2023 by v_Enhance
Want to help AoPS remain a valuable Olympiad resource? Help us add contests to AoPS's Contest Collections.

Find instructions and a list of contests to add here: https://artofproblemsolving.com/community/c40244h1064480_contests_to_add
1 reply
dcouchman
Sep 9, 2019
v_Enhance
Apr 5, 2023
k i Zero tolerance
ZetaX   49
N May 4, 2019 by NoDealsHere
Source: Use your common sense! (enough is enough)
Some users don't want to learn, some other simply ignore advises.
But please follow the following guideline:


To make it short: ALWAYS USE YOUR COMMON SENSE IF POSTING!
If you don't have common sense, don't post.


More specifically:

For new threads:


a) Good, meaningful title:
The title has to say what the problem is about in best way possible.
If that title occured already, it's definitely bad. And contest names aren't good either.
That's in fact a requirement for being able to search old problems.

Examples:
Bad titles:
- "Hard"/"Medium"/"Easy" (if you find it so cool how hard/easy it is, tell it in the post and use a title that tells us the problem)
- "Number Theory" (hey guy, guess why this forum's named that way¿ and is it the only such problem on earth¿)
- "Fibonacci" (there are millions of Fibonacci problems out there, all posted and named the same...)
- "Chinese TST 2003" (does this say anything about the problem¿)
Good titles:
- "On divisors of a³+2b³+4c³-6abc"
- "Number of solutions to x²+y²=6z²"
- "Fibonacci numbers are never squares"


b) Use search function:
Before posting a "new" problem spend at least two, better five, minutes to look if this problem was posted before. If it was, don't repost it. If you have anything important to say on topic, post it in one of the older threads.
If the thread is locked cause of this, use search function.

Update (by Amir Hossein). The best way to search for two keywords in AoPS is to input
[code]+"first keyword" +"second keyword"[/code]
so that any post containing both strings "first word" and "second form".


c) Good problem statement:
Some recent really bad post was:
[quote]$lim_{n\to 1}^{+\infty}\frac{1}{n}-lnn$[/quote]
It contains no question and no answer.
If you do this, too, you are on the best way to get your thread deleted. Write everything clearly, define where your variables come from (and define the "natural" numbers if used). Additionally read your post at least twice before submitting. After you sent it, read it again and use the Edit-Button if necessary to correct errors.


For answers to already existing threads:


d) Of any interest and with content:
Don't post things that are more trivial than completely obvious. For example, if the question is to solve $x^{3}+y^{3}=z^{3}$, do not answer with "$x=y=z=0$ is a solution" only. Either you post any kind of proof or at least something unexpected (like "$x=1337, y=481, z=42$ is the smallest solution). Someone that does not see that $x=y=z=0$ is a solution of the above without your post is completely wrong here, this is an IMO-level forum.
Similar, posting "I have solved this problem" but not posting anything else is not welcome; it even looks that you just want to show off what a genius you are.

e) Well written and checked answers:
Like c) for new threads, check your solutions at least twice for mistakes. And after sending, read it again and use the Edit-Button if necessary to correct errors.



To repeat it: ALWAYS USE YOUR COMMON SENSE IF POSTING!


Everything definitely out of range of common sense will be locked or deleted (exept for new users having less than about 42 posts, they are newbies and need/get some time to learn).

The above rules will be applied from next monday (5. march of 2007).
Feel free to discuss on this here.
49 replies
ZetaX
Feb 27, 2007
NoDealsHere
May 4, 2019
Weird Geo
Anto0110   0
3 minutes ago
In a trapezium $ABCD$, the sides $AB$ and $CD$ are parallel and the angles $\angle ABC$ and $\angle BAD$ are acute. Show that it is possible to divide the triangle $ABC$ into 4 disjoint triangle $X_1. . . , X_4$ and the triangle $ABD$ into 4 disjoint triangles $Y_1,. . . , Y_4$ such that the triangles $X_i$ and $Y_i$ are congruent for all $i$.
0 replies
Anto0110
3 minutes ago
0 replies
Is the geometric function injective?
Project_Donkey_into_M4   1
N 24 minutes ago by Funcshun840
Source: Mock RMO TDP and Kayak 2018, P3
A non-degenerate triangle $\Delta ABC$ is given in the plane, let $S$ be the set of points which lie strictly inside it. Also let $\mathfrak{C}$ be the set of circles in the plane. For a point $P \in S$, let $A_P, B_P, C_P$ be the reflection of $P$ in sides $\overline{BC}, \overline{CA}, \overline{AB}$ respectively. Define a function $\omega: S \rightarrow \mathfrak{C}$ such that $\omega(P)$ is the circumcircle of $A_PB_PC_P$. Is $\omega$ injective?

Note: The function $\omega$ is called injective if for any $P, Q \in S$, $\omega(P) = \omega(Q) \Leftrightarrow P = Q$
1 reply
Project_Donkey_into_M4
3 hours ago
Funcshun840
24 minutes ago
numbers at vertices of triangle / tetrahedron, consecutive and gcd related
parmenides51   1
N 44 minutes ago by TheBaiano
Source: 2022 May Olympiad L2 p4
a) A positive integer is written at each vertex of a triangle. Then on each side of the triangle the greatest common divisor of its ends is written. It is possible that the numbers written on the sides be three consecutive integers, in some order?
b) A positive integer is written at each vertex of a tetrahedron. Then, on each edge of the tetrahedron is written the greatest common divisor of its ends . It is possible that the numbers written in the edges are six consecutive integers, in some order?
1 reply
parmenides51
Sep 4, 2022
TheBaiano
44 minutes ago
red squares in a 7x7 board
parmenides51   2
N an hour ago by TheBaiano
Source: 2022 May Olympiad L2 p1
In a $7\times7$ board, some squares are painted red. Let $a$ be the number of rows that have an odd number of red squares and let $b$ be the number of columns that have an odd number of red squares. Find all possible values of $a+b$. For each value found, give a example of how the board can be painted.
2 replies
parmenides51
Sep 4, 2022
TheBaiano
an hour ago
winning strategy, vertices of regular n-gon
parmenides51   1
N an hour ago by TheBaiano
Source: 2022 May Olympiad L2 p5
The vertices of a regular polygon with $N$ sides are marked on the blackboard. Ana and Beto play alternately, Ana begins. Each player, in turn, must do the following:
$\bullet$ join two vertices with a segment, without cutting another already marked segment; or
$\bullet$ delete a vertex that does not belong to any marked segment.
The player who cannot take any action on his turn loses the game. Determine which of the two players can guarantee victory:
a) if $N=28$
b) if $N=29$
1 reply
parmenides51
Sep 4, 2022
TheBaiano
an hour ago
another functional inequality?
Scilyse   31
N an hour ago by Andyexists
Source: 2023 ISL A4
Let $\mathbb R_{>0}$ be the set of positive real numbers. Determine all functions $f \colon \mathbb R_{>0} \to \mathbb R_{>0}$ such that \[x \big(f(x) + f(y)\big) \geqslant \big(f(f(x)) + y\big) f(y)\]for every $x, y \in \mathbb R_{>0}$.
31 replies
Scilyse
Jul 17, 2024
Andyexists
an hour ago
Apple sharing in Iran
mojyla222   2
N an hour ago by sami1618
Source: Iran 2025 second round p6
Ali is hosting a large party. Together with his $n-1$ friends, $n$ people are seated around a circular table in a fixed order. Ali places $n$ apples for serving directly in front of himself and wants to distribute them among everyone. Since Ali and his friends dislike eating alone and won't start unless everyone receives an apple at the same time, in each step, each person who has at least one apple passes one apple to the first person to their right who doesn't have an apple (in the clockwise direction).

Find all values of $n$ such that after some number of steps, the situation reaches a point where each person has exactly one apple.
2 replies
1 viewing
mojyla222
Today at 4:17 AM
sami1618
an hour ago
confusing inequality
giangtruong13   5
N 2 hours ago by arqady
Let $a,b,c>0$ such that: $a^2b^2+ c^2b^2+ a^2c^2=3(abc)^2$. Prove that: $$\sum \frac{b+c}{a} \geq 2\sqrt{3(ab+bc+ca)}$$
5 replies
giangtruong13
Apr 18, 2025
arqady
2 hours ago
Orthoincentre mixup in rmo mock
Project_Donkey_into_M4   1
N 2 hours ago by RANDOM__USER
Source: Mock RMO 2018,TDP and Kayak P5
Let $\Delta ABC$ be a triangle with circumcircle $\omega$, $P_A, P_B, P_C$ be the foot of altitudes from $A, B, C$ onto the opposite sides respectively and $H$ the orthocentre. Reflect $H$ across the line $BC$ to obtain $Q$. Suppose there exists points $I,J \in \omega$ such that $P_A$ is the incentre of $\Delta QIJ$. If $M$ and $N$ be the midpoints of $\overline{P_AP_B}$ and $\overline{P_AP_C}$ respectively, then show that $I,J,M,N$ are collinear.
1 reply
Project_Donkey_into_M4
3 hours ago
RANDOM__USER
2 hours ago
Advanced topics in Inequalities
va2010   17
N 2 hours ago by Novmath
So a while ago, I compiled some tricks on inequalities. You are welcome to post solutions below!
17 replies
va2010
Mar 7, 2015
Novmath
2 hours ago
Estonian Math Competitions 2005/2006
STARS   1
N 2 hours ago by Rohit-2006
Source: Juniors Problem 4
A $ 9 \times 9$ square is divided into unit squares. Is it possible to fill each unit square with a number $ 1, 2,..., 9$ in such a way that, whenever one places the tile so that it fully covers nine unit squares, the tile will cover nine different numbers?
1 reply
STARS
Jul 30, 2008
Rohit-2006
2 hours ago
TST lavil combo endup in scariness
Project_Donkey_into_M4   0
3 hours ago
Source: Mock RMO TDP and Kayak 2018 P6
Let $n \geq 2$ be a positive integer. There're $n$ roads in Mahismati, no three roads are concurrent and no two roads are parallel (so each two of them intersect). Number all possible $4\binom{n}{2}$ angles formed by intersection points with $\{1, 2, \cdots, 4\binom{n}{2} \}$ in any order.

Baahubali moves in the road in the following way: he starts at an point in the road, moves forward until he reaches an intersection and every time he meets an intersection, he moves either left or right, alternating his choice at each intersection point.

Now, for a path traced by Baahubali, we say the path touches a numbered angle $\angle (\ell_1, \ell_2)$ if he goes through $\ell_1$, turns at the intersection point around the angle and then continues to $\ell_2$ (or vice versa). Katappa colors two numbers $i,j$ with $1 \leq i, j \leq 4\binom{n}{2}$ with the same color iff there's a path traced by Baahubali which touches the angles numbered $i,j$ simultaneously.

Is it true that atleast $2n$ colors will be used by Katappa to color all the numbers?

Example for $n = 3$ : In the picture attached below, total $2 \times 3 = 6$ colors will be used to color the numbers in each set with same color $\{1 \}, \{6 \}, \{11 \}, \{4, 9, 7 \}, \{12, 3, 5 \}, \{10, 8, 2 \}$. The numbers $\{4, 9, 7 \}$ will be colored with the same color because the red path traced by Baahubali touches the angles $4, 9, 7$.

https://i.stack.imgur.com/89KpO.png
0 replies
Project_Donkey_into_M4
3 hours ago
0 replies
standard Q FE
jasperE3   0
3 hours ago
Source: gghx, p19004309
Find all functions $f:\mathbb Q\to\mathbb Q$ such that for any $x,y\in\mathbb Q$:
$$f(xf(x)+f(x+2y))=f(x)^2+f(y)+y.$$
0 replies
jasperE3
3 hours ago
0 replies
Get golden ratio-ed
Project_Donkey_into_M4   0
3 hours ago
Source: Mock RMO TDP and Kayak 2018,P4
Let $\phi = \dfrac{1+\sqrt{5}}{2}$, and $a$ be a positive integer. and $f(a)$ denote the number of solutions to $a \lceil b \phi \rceil - b \lfloor a \phi \rfloor = 1$, where $b$ is allowed to vary on the set of positive integers
Prove there's a constant $c$ such that $0 \leq f(a) \leq c$ for all $a$, and $f(a) = c$ infintiely many often.
Let $S = \{\lceil \log_2 a \rceil + 1 | f(a) = c, a \in \mathbb{N} \}$. Prove there are infinitely many primes $p$ such that $p$ divides atleast one element of $S$
0 replies
Project_Donkey_into_M4
3 hours ago
0 replies
IMO Problem 5
iandrei   23
N Apr 15, 2025 by eevee9406
Source: IMO ShortList 2003, algebra problem 4
Let $n$ be a positive integer and let $x_1\le x_2\le\cdots\le x_n$ be real numbers.
Prove that

\[
          \left(\sum_{i,j=1}^{n}|x_i-x_j|\right)^2\le\frac{2(n^2-1)}{3}\sum_{i,j=1}^{n}(x_i-x_j)^2.
             \]
Show that the equality holds if and only if $x_1, \ldots, x_n$ is an arithmetic sequence.
23 replies
iandrei
Jul 14, 2003
eevee9406
Apr 15, 2025
IMO Problem 5
G H J
Source: IMO ShortList 2003, algebra problem 4
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
iandrei
138 posts
#1 • 6 Y
Y by bel.jad5, Adventure10, mathematicsy, Mango247, MS_asdfgzxcvb, radian_51
Let $n$ be a positive integer and let $x_1\le x_2\le\cdots\le x_n$ be real numbers.
Prove that

\[
          \left(\sum_{i,j=1}^{n}|x_i-x_j|\right)^2\le\frac{2(n^2-1)}{3}\sum_{i,j=1}^{n}(x_i-x_j)^2.
             \]
Show that the equality holds if and only if $x_1, \ldots, x_n$ is an arithmetic sequence.
Attachments:
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Anonymous
334 posts
#2 • 4 Y
Y by iceillusion, Adventure10, Mango247, radian_51
We can rewrite the given expression as
\( \sum_i (2i-n-1)x_i \)^2
\leq (n^2 - 1)/3 \( n\sum x_i^2 - (\sum x_i )^2\).
As the expression is translation invariant (obvious from the original form), we can assume that \sum x_i =0. The rest follows by Cauchy-Schwartz.
Z K
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Anonymous
334 posts
#3 • 3 Y
Y by Adventure10, Mango247, radian_51
The crucial idea is that we may assume that Sum x_k = 0. I had tried to solve this problem, but I did not succeed. I came up with the first identity in Jayanta's message, but I was not smart enough to make his clever assumption. Let me complete the proof

(a) Since sum_{i,j} |x_j x_i|= 2 sum_{i<j} (x_j x_i) and sum_{i,j} (x_j x_i)^2 = 2 sum{i<j} (x_j x_i)^2 , it suffices to show that

( sum_{i<j} (x_j x_i) )^2 <= [(n^2-1)/3] sum{i<j} (x_j x_i)^2.

Since
sum_{i<j} (x_j x_i)^2 = n sum x_i^2 (sum x_i )2 ,

and assuming that sum x_i =0, it suffices to show that

( sum_{i<j} (x_j x_i) )^2 <= [n(n^2-1)/3] sum x_i^2

We observe that

(*) sum_{i<j} (x_j x_i)= (n-1) x_n +(n-3) x_{n-1} +-(n-3)x_2-(n-1) x_1=sum a_k x_k,

where a_{n-k} = n-1-2k, for k=1,,,n. Note that if n=2m, then

sum a_k^2 = 2(1^2+3^2++(2m-1)^2 ) = 2m(2m-1)(2m+1)/3 = n(n^2-1)/3,

and if n=2m-1, then a_{n-k} = 2(m-k), for k=1,,n (note that a_m=0), so

sum a_k^2 =8[(1^2+2^2++(m-1)^2] = 8(m-1)m(2m-1)/6 = n(n^2-1)/3,

since m(m-1)=(n^2-1)/4. Thus, by Cauchy's inequality we have,

( sum_{i<j| (x_j x_i) )2 <= (sum a_k^2 ) (sum x_k^2 ) ={ n(n^2-1)/3}(sum x_k^2 ),

as desired.

Comment: To see (*), consider the following triangle

x_n-x_1 x_{n-1}-x_1 x_3-x_1 x_2-x_1
x_n-x_2 x_{n-1}-x_2 x_3-x_2

x_{n-1}-x_{n-2}
x_n-x_{n-2}
x_n-x_{n-1}

(b) We have quality of if and only if x_k= 0 for all k, or a_k= t*x_k for all k and some nonzero real number t. In the first case clearly we have an arithmetic progression. In the second case, since 2 = a_k-a_{k-1} = t(x_k-x_{k-1}), we have x_k = x_{k-1} + 2/t, and the result follows.
Z K
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Anonymous
334 posts
#4 • 3 Y
Y by Adventure10, Mango247, radian_51
well, I meant

"Since
sum_{i<j} (x_j x_i)^2 = n sum x_i^2 (sum x_i )^2 ,

and assuming that sum x_i =0, .....e.t.c"
Z K
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
galois
400 posts
#5 • 3 Y
Y by Adventure10, Mango247, radian_51
this one was a rather cute problem.but two people have beaten me agen b4 i cud post my solution.and my proof is same as achilleas proof(u guys are fast)
which wud u ppl rate as the best question at this year's imo(i luvd no2)
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Anonymous
334 posts
#6 • 3 Y
Y by Adventure10, Mango247, radian_51
Question: Is the following inequality true?

Let n>2 and x_1 <=x_2<=...<=x_n and y_1<=y_2<=...<=y_n. Then

(sum_{i,j} |x_i-x_j|) * (sum_{i,j} |y_i-y_j|) <= 2(n^2-1)/3 sum_{i,j} |x_i-x_j)(y_i-y_j)| .

This would be a generalization of the IMO problem.


We could assume that sum x_k=sum y_k=0 and use the identity

n*(sum x_k * y_k) = (sum x_k)*(sum y_k) + sum_{i<j} (x_i-x_j)(y_i-y_j).

Then it would suffice to show that

(sum_{i<j} (x_j-x_j)) * (sum_{i<j} (y_j-y_i) ) <= [2n(n^2-1)/3] sum x_k * y_k


Since sum_{i<j} (x_j-x_i)) = sum a_k * x_k and sum_{i<j} (y_j-y_i)) = sum a_k * y_k, where a_k are as above for k=1,2,..,n-1 (I had written up to n inadvertently in my previous message).
Also sum (a_k)^2=2n(n^2-1)/3 (proof in my previous message) and

sum_{i<j} (x_j-x_j)) * (sum_{i<j} (y_j-y_i) ) =(sum a_k * x_k) * (sum a_k * y_k)

but I cannot see how this would prove it; it seems like a deadend. Any other approach? (Of course, the inequality could be false)
Z K
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Anonymous
334 posts
#7 • 3 Y
Y by Adventure10, Mango247, radian_51
let a_k=x_{k+1}-x_k>=0, then the rhs (except the (n^2-1)/3 factor) is
[a_1^2+a_2^2...+a_{n-1}^2 +
(a_1+a_2)^2+(a_2+a_3)^2+...+(a_{n-2}+a_{n-1})^2+
...
+(a_1+a_2+...+a_{n-1})^2]
write a k-sum square as
(a_{i+1}+...+a_{i+k})^2 as k^2 [(a_{i+1}+...+a_{i+k})/k]^2, so altogether we have
(n-1)*1^2+(n-2)*2^2+...+1*(n-1)^2 terms, then use the power mean inequality to get it
cheers,
Z K
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Anonymous
334 posts
#8 • 3 Y
Y by Adventure10, Mango247, radian_51
Elinor, I hope this is the way you did it..

The number of terms is

(n-1)*1^2+(n-2)*2^2+...+1*(n-1)^2 = sum (n-k)*k^2 for k=1,2,...,n-1 = n^2(n^2-1)/12.

So the rhs of the reduced inequality (i.e. the sum for i<j) times the numer of terms n^2(n^2-1)/12 is larger than or equal to the square of the sum

S=(a_1+...+a_{n-1} +2*(a_1+a_2) + 2*(a_2+a_3)+...+2*(a_{n-2}+a_{n-1})+..+(n-1)*(a_1+....+a_{n-1}))

(This also follows from Cauchy-Scwartz). In order to finish the proof we need to show that

S>= (n/2)*S' , where S' is the sum=(a_1+...+a_{n-1} +
(a_1+a_2) + (a_2+a_3)+...+(a_{n-2}+a_{n-1})+..+(a_1+....+a_{n-1}))

In order to do this we observe that after making the operations in the two sums we see that for k=1,2,..,n-1, the coefficient of a_k in S is nk(n-k)/2 and in S' is k(n-k). (of course, this needs proof!!) Thus we actually have equality

S= (n/2)*S' ,

and the result follows.
Z K
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Alison
264 posts
#9 • 2 Y
Y by Adventure10, Mango247
Achilleas Sinefakopoulos wrote:
Question: Is the following inequality true?

Let n>2 and x_1 <=x_2<=...<=x_n and y_1<=y_2<=...<=y_n. Then

(sum_{i,j} |x_i-x_j|) * (sum_{i,j} |y_i-y_j|) <= 2(n^2-1)/3 sum_{i,j} |x_i-x_j)(y_i-y_j)| .

...(Of course, the inequality could be false)

It is false in general, (as will be shown later). This can also be easily seen by trying n=3. However under certain restrictions for the x_i, y_i the attack given by Elinor and Achilles can work by replacing the Cauchy-Schwartz (or AM-RMS) inequality with Chebyshev's inequality. However this gives the very awkward restriction that if f(i, j) = |x_i-x_j|/|i-j| and g(i,j) = |y_i-y_j|/|i-j|, f(i, j)<f(k,l) iff g(i, j)<g(k,l). I can't think of any interesting pairs of sequences that satisfy this condition, other than y_i = x_i + ki for k in R.

Also, there is a corresponding inequality using the other form of Chebyshev if f(i, j)<f(k,l) iff g(i, j)>g(k,l), showing that the originally proposed inequality is in general false.
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Anonymous
334 posts
#10 • 3 Y
Y by Adventure10, Mango247, radian_51
Alison, thanks for your reply!
Z K
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Zhero
2043 posts
#11 • 6 Y
Y by Leooooo, karitoshi, iceillusion, Adventure10, ehuseyinyigit, radian_51
Lemma: $\sum_{i,j=1}^n |x_i - x_j| |i - j| = \frac{n \sum_{i,j=1} |x_i - x_j|}{2}$.
Proof: Let $a_1 = x_1$, and let $a_{i} = x_i - x_{i-1}$ for all $i$ with $2 \leq i \leq n$. The lemma we wish to prove can easily be seen to be equivalent to
$\sum_{1 \leq i < j \leq n} (i - j)(a_{i+1} + a_{i+2} + \cdots + a_j) = \frac{n}{2} \sum_{1 \leq i < j \leq n} (a_{i+1} + a_{i+2} + \cdots + a_j).$

When expanded, both sides are a linear combination of $a_1, a_2, \cdots, a_n$. We will determine the coefficients of each variable on both sides of the equation.

On the left-hand side, $a_k$ appears if and only if $i+1 \leq k \leq j$, in which case it comes with a factor if $(j-i)$. Hence, the coefficient of $a_k$ on the left-hand side is $\sum_{i=1}^{k-1} \sum_{j=k}^n (j-i) = \frac{n(k-1)(n-k+1)}{2}$. On the right-hand side, $a_k$ appears iff $i+1 \leq k \leq j$, in which case it comes with a factor if $\frac{n}{2}$. Hence, the coefficient of $a_k$ on the right-hand side is $\frac{n(k-1)(n-k+1)}{2}$, which is equal to the coefficient on the left. $\blacksquare$

It can easily be seen that $\sum_{i=1}^n \sum_{j=1}^n |i - j|^2 = \frac{2(n^2-1)}{3} \cdot \frac{n^2}{4}$. By the Cauchy-Schwarz inequality,
\begin{align*}
\frac{2(n^2 - 1)}{3} \left( \sum_{i,j=1}^n |x_i - x_j|^2 \right)
&= \frac{\left( \sum_{i,j=1}^n |x_i - x_j|^2 \right) \left( \sum_{j=1}^n |i-j|^2 \right)}{\frac{n^2}{4}} \\
&\geq \left (\frac{ \sum_{i,j=1}^n |x_i - x_j| \, |i - j| }{ \frac{n}{2} } \right)^2 \\
&= \left( \sum_{i,j=1}^n |x_i - x_j| \right)^2, 
\end{align*}
as desired.

The cases of equality easily follow from the above application of Cauchy-Schwarz.
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Swistak
180 posts
#12 • 2 Y
Y by Adventure10, Mango247
I have known that this problem can be solved using Cauchy-Schwarz inequality, but even with this information I haven't known how to solve it. It can be solved by using derivatives in a very simple way. Just compute derivatives on each variable and it can be easily seen that $x_2-x_1=x_3-x_2$ and so on.
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
superpi83
1416 posts
#13 • 6 Y
Y by iceillusion, Vladimir_Djurica, Adventure10, Mango247, MS_asdfgzxcvb, radian_51
We first make use of symmetry to rewrite the inequality as
\[\left(\sum_{1\le i<j\le n}|x_i-x_j|\right)^2\le\frac{n^2-1}3\left(\sum_{1\le i<j\le n}|x_i-x_j|^2\right)\].
WLOG that $x_1\le x_2\le\dots\le x_n$ and let $x_{i-1}-x_i=a_i$. The inequality is equivalent to
\[\left(\sum_{1\le i<j\le n}\left(a_i+\dots+a_{j-1}\right) \right)^2\le\frac{n^2-1}3\left(\sum_{1\le i<j\le n}\left(a_i+\dots+a_{j-1}\right)^2\right)\] for all $a_1,\dots,a_{n-1}$.
But this can be rewritten as
\[\left(\sum_{l=1}^{n-1}\sum_{j-i=l}\left(a_i+\dots+a_{j}\right) \right)^2\le\frac{n^2-1}3\left(\sum_{l=1}^{n-1}\sum_{j-i=l}\left(a_i+\dots+a_{j}\right)^2\right)\]
By Cauchy-Schwarz:
\begin{align*}
\left(\sum_{l=1}^{n-1}\sum_{j-i=l}\left(a_i+\dots+a_{j}\right)^2\right)\left(\sum_{l=1}^{n-1}\sum_{j-i=l}l^2\right)&\ge\left(\sum_{l=1}^{n-1}\sum_{j-i=l}l\left(a_i+\dots+a_j\right)\right)^2\\
&=\left(\sum_{l=1}^{n-1}l\sum_{j-i=l}\left(a_i+\dots+a_j\right)\right)^2
\end{align*}

We claim that
\[\sum_{j-i=l}(a_i+\dots+a_j)=\sum_{j-i=(n-l)}(a_i+\dots+a_j)\]. Indeed, we may consider the $l\times(n-l)$ matrix:
\[ \left( \begin{array}{cccc}
a_1 & a_2 & \dots & a_l \\
a_2 & a_3 & \dots & a_{l+1} \\
\vdots & \vdots & \ddots & \vdots\\
a_{n-l} & a_{n-l+1} & \dots & a_n \end{array} \right)\]
The first sum corresponds to summing the matrix row by row, and the second corresponds to summing it column by column. Thus the two sums are equal, as claimed.

Hence:
\begin{align*}
\left(\sum_{l=1}^{n-1}l\sum_{j-i=l}\left(a_i+\dots+a_j\right)\right)^2&=\left(\sum_{l=1}^{n-1}\frac n2\sum_{j-i=l}\left(a_i+\dots+a_j\right)\right)^2\\
&=\frac{n^2}4\left(\sum_{l=1}^{n-1}\sum_{j-i=l}\left(a_i+\dots+a_j\right)\right)^2
\end{align*}

We may also check that
\[\sum_{l=1}^{n-1}\sum_{j-i=l}l^2=\sum_{l=1}^{n-1}(n-l)l^2=\frac{n^4-n^2}{12}\]. Thus we have proven that
\[\frac{n^4-n^2}{12}\left(\sum_{l=1}^{n-1}\sum_{j-i=l}\left(a_i+\dots+a_{j}\right)^2\right)\ge\frac{n^2}4\left(\sum_{l=1}^{n-1}\sum_{j-i=l}\left(a_i+\dots+a_j\right)\right)^2\]
Dividing $\frac{n^2}4$ yields
\[\frac{n^2-1}{3}\left(\sum_{l=1}^{n-1}\sum_{j-i=l}\left(a_i+\dots+a_{j}\right)^2\right)\ge\left(\sum_{l=1}^{n-1}\sum_{j-i=l}\left(a_i+\dots+a_j\right)\right)^2\] as desired.

Furthermore, from Cauchy's equality condition, equality holds if and only if $a_1=a_2=\dots=a_{n-1}$ - that is, when the $x_i$ form an arithmetic sequence.
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
bel.jad5
3750 posts
#15 • 2 Y
Y by Adventure10, solasky
This is a very old inequality and there are a number of elegant solutions but I would like to contribute with a very straightforward approach using induction and derivatives.
For $n=3$, the inequality is obviously true. Now let's assume that the inequality is true for $n$ and we will prove it for $n+1$.
Let's define $y_{i}$ such that $y_{i}=x_{i}-\frac{\sum_{j=1}^{n}x_{j}}{n}$ and we have $\sum_{i=1}^{n} y_{i}=0$
We need to prove that for $y_{n+1} \geq y_{n}$ we have: $P(y_{n+1}) \geq 0$ where:
\[P(y_{n+1}) = \frac{2}{3}((n+1)^2-1)(\sum_{i,j=1}^{n+1} (y_{i}-y_{j})^2) - (\sum_{i,j=1}^{n+1} |y_{i}-y_{j}|)^2 \geq 0\]Let $A_{n} = \sum_{i,j=1}^{n} (y_{i}-y_{j})^2$ and $B_{n} = \sum_{i,j=1}^{n} |y_{i}-y_{j}|$ and we have from induction's hypothesis $\frac{2}{3}(n^2-1)A_{n} \geq B_{n}^2 $.
After Simplification, we get:
\[P(y_{n+1}) = \frac{2}{3}((n+1)^2-1)(2ny_{n+1}^2+\frac{n+1}{n}A_{n})-(2ny_{n+1}+B_{n})^2\]$P$ reaches a minimal value for $y_{n+1}$ such that $P'(y_{n+1})=0$ and after simplification, we get:
\[P'(y_{n+1})=0 \Leftrightarrow	 y_{n+1}=\frac{3B_{n}}{2n(n-1)}\]Then we have:
\[P(y_{n+1}) \geq P(\frac{3B_{n}}{2n(n-1)})=\frac{n+2}{n-1}(\frac{2}{3}(n^2-1)A_{n}-B_{n}^2) \geq 0\]Equality case: we have $\frac{2}{3}(n^2-1)A_{n}-B_{n}^2 = 0$ which is true only if there exists a real number $r$ such that $x_{i}=ir$ for all $i\leq n$, and
\[x_{n+1}=y_{n+1}+\frac{\sum_{i=1}^{n} x_{i}}{n}=\frac{3\sum_{i,j=1}^{n}|i-j|r}{2n(n-1)}+\frac{\sum_{i=1}^{n} ir}{n}=\frac{(n+1)r}{2}+\frac{(n+1)r}{2}=(n+1)r\]Which completes the proof by induction.
This post has been edited 2 times. Last edited by bel.jad5, Jul 23, 2017, 11:29 AM
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
illogical_21
1034 posts
#16 • 5 Y
Y by PIartist, Zfn.nom-_nom, Adventure10, Mango247, radian_51
basically the same as first cauchy sol

WLOG let $x_1\le x_2\le \cdots \le x_n$, then we work on the LHS,
\begin{align*}
\sum_{i=1}^n \sum_{j=1}^n |x_i-x_j|
&=\sum_{i=1}^n \left(\sum_{j=1}^{i-1}(x_i-x_j)+\sum_{j=i+1}^{n}(x_j-x_i)\right)\\
&=\sum_{i=1}^n(2i-n-1)x_i+\sum_{i=1}^n \left(\sum_{j=i+1}^{n}x_j-\sum_{j=1}^{i-1}x_j\right)\\
&=2\sum_{i=1}^n(2i-n-1)x_i.
\end{align*}Now we work on the RHS,
\begin{align*}
\sum_{i=1}^n \sum_{j=1}^n |x_i-x_j|^2
&=\sum_{i=1}^n\sum_{j=1}^n (x_i^2+x_j^2)-2\sum_{i=1}^n\sum_{j=1}^n x_ix_j\\
&=2n\sum_{i=1}^n x_i^2-2\left(\sum_{i=1}^n x_i\right)^2.\\
\end{align*}
The key idea is that the original inequality does not depend on $\sum_{i=1}^n x_i$, so we may set it equal to zero. Then the desired inequality becomes
\[\left(\sum_{i=1}^n(2i-n-1)x_i\right)^2\le\frac{n(n^2-1)}{3}\sum_{i=1}^n x_i^2.\]
Applying C-S on the LHS, we have
\[\left(\sum_{i=1}^n(2i-n-1)x_i\right)^2\le\left(\sum_{i=1}^n (2i-n-1)^2\right)\left(\sum_{i=1}^n x_i^2\right),\]so it suffices to show that
\[\sum_{i=1}^n (2i-n-1)^2=\frac{n(n^2-1)}{3}.\]
If $n=2m$, we have
\begin{align*}
\sum_{i=1}^n (2i-n-1)^2 &= 2(1^2+3^2+\cdots+(2m-1)^2)\\
&=2\sum_{i=1}^{m}\left(\binom{2i-1}{2}+\binom{2i}{2}\right)\\
&=2\binom{2m+1}{3}\\
&=\frac{n(n^2-1)}{3},
\end{align*}as desired.

If $n=2m+1$, we have
\begin{align*}
\sum_{i=1}^n (2i-n-1)^2 &= 2(2^2+\cdots+(2m)^2)\\
&=2\sum_{i=1}^{m}\left(\binom{2i}{2}+\binom{2i+1}{2}\right)\\
&=2\binom{2m+2}{3}\\
&=\frac{n(n^2-1)}{3},
\end{align*}as desired.

The equality case occurs at
\[x_1:x_2:\cdots:x_n=1-n:3-n:\cdots:n-1,\]so they are in arithmetic progression, as desired.
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
IndoMathXdZ
691 posts
#17 • 3 Y
Y by karitoshi, Adventure10, radian_51
This would have been a really nice inequality problem if the question hadn't stated the inequality case.
Anyway, posting this for storage, basically the same CS solution as others.
We'll first rewrite the sum condition as
\[ \left( \sum_{1 \le i < j \le n} (x_j - x_i) \right)^2 \le \frac{n^2 - 1}{3} \sum_{1 \le i < j \le n} (x_j - x_i)^2 \]Now, we'll prove that
\[ \sum_{1 \le i < j \le n} \frac{(j - i)^2}{n^2} = \frac{n^2}{12} \]This can be proved by noticing that
\begin{align*}
 \sum_{1 \le i < j \le n} \frac{(j - i)^2}{n^2} &= \frac{1}{n^2} \sum_{1 \le i <j \le n} (j - i)^2 \\
&= \frac{1}{n^2} \sum_{k = 1}^{n-1} k^2 (n - k) \\
&= \frac{1}{n} \sum_{k = 1}^{n-1} k^2 - \frac{1}{n^2} \sum_{k = 1}^{n-1} k^3 \\
&= \frac{1}{n} \left( \frac{1}{6} (n-1)(n)(2n-1) \right) - \frac{1}{n^2} \left( \frac{(n-1)n}{2} \right)^2 \\
&= \frac{(n-1)(2n-1)}{6} - \frac{(n-1)^2}{4} \\
&= \frac{2(2n^2 - 3n + 1) - 3(n^2 - 2n + 1)}{12} \\
&= \frac{n^2 - 1}{12}
\end{align*}which is what we wanted.
Therefore, we could rewrite the condition as
\[ \left( \sum_{1 \le i < j \le n} (x_j - x_i) \right)^2 \le 4 \left( \sum_{1 \le i < j \le n} \frac{(j - i)^2}{n^2} \right) \left( \sum_{1 \le i < j \le n} (x_j - x_i)^2 \right) \]Now, apply CS at the RHS, we have
\[ 4 \left( \sum_{1 \le i < j \le n} \frac{(j - i)^2}{n^2} \right) \left( \sum_{1 \le i < j \le n} (x_j - x_i)^2 \right) \ge \left( 2 \sum_{1 \le i < j \le n} \frac{j - i}{n} (x_j - x_i) \right)^2 \]It suffices to prove that
\[ 2 \sum_{1 \le i < j \le n} \frac{j - i}{n} (x_j - x_i) =  \sum_{1 \le i < j \le n} (x_j - x_i) \]But this is true by considering each coefficient of $x_i$.
Notice that in the RHS, the coefficient of $x_j$ is $( j - 1) - (n - j) = 2j - n - 1$ and in the LHS, we have the coefficient of $x_j$ is
\[ \frac{2 ( (1 + 2 + \dots + j - 1) - (1 + \dots + (n - j)) )}{n} = \frac{(j - 1)j - (n -j)(n-j+1)}{n} = \frac{2nj - n^2 - n}{n} = 2j - n - 1\]
This post has been edited 1 time. Last edited by IndoMathXdZ, Nov 12, 2019, 9:24 AM
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
jeteagle
480 posts
#18
Y by
Let $L$ and $R$ equal the LHS and RHS of the inequality respectively. We have $R = \frac{4(n^2-1)}{3}\sum_{i>j\ge1}^{n}(x_i-x_j)^2$ since $(x_i-x_j)^2 = (x_j-x_i)^2$. Next, let $J = \sum_{i>j\ge1}^{n}(x_i-x_j)^2$ and $K = \sum_{i = 1}^n\sum_{i \ge j \ge 1}^nx_j^2$. Also, let $c_n = \binom{n}{2}, c_{n-1} = \binom{n}{2}-n, \ldots c_1 = \binom{n}{2}-(n-1)n = -\binom{n}{2}$. By Cauchy-Schwarz, $$JK = \left((x_2-x_1)^2+(x_3-x_1)^2+(x_3-x_2)^2+\ldots+(x_n^2-x_{n-1}^2)\right)\left((1^2)+(2^2+1^2)+(3^2+\ldots+2^2+1^2)\right)$$$$\ge \left(1(x_2-x_1)+2(x_3-x_1)+1(x_3-x_2)+\ldots+2(x_n-x_{n-2})+1(x_n-x_{n-1})\right)^2$$$$= \left(c_nx_n+c_{n-1}x_{n-1}+c_{n-2}x_{n-2}+\ldots+c_1x_1\right)^2 = I^2$$where the last step can be proven by considering specific values of $x_i$.

Next, I will find the specific value of $K$. We have the following lemmas: $\sum_{i = 1}^ni^3 = \frac{n^2(n+1)^2}{4}, \sum_{i = 1}^ni^2 = \frac{n(n+1)(2n+1)}{6}, \sum_{i = 1}^ni = \frac{n(n+1)}{2}.$ Therefore, $$K = \sum_{i = 1}^{n-1} \frac{i(i+1)(2i+1)}{6} = \sum_{i = 1}^n \frac{2i^3+3i^2+i}{6} = \frac{\frac{(n-1)^2n^2}{2}+\frac{(n-1)n(2n-1)}{2}+\frac{(n-1)n}{2}}{6} = \frac{(n-1)n^2(n+1)}{12}.$$Therefore, we have $$R = \frac{4(n^2-1)}{3K}JK \ge \frac{4(n-1)(n+1)}{\frac{3(n-1)n^2(n+1)}{12}}I^2 = \frac{16}{n^2}I^2.$$
Finally, it suffices to prove $L = \frac{16}{n^2}I^2$. Note that $L = 4\left(\sum_{i \ge 1 \ge 1}|x_i-x_j|\right)^2$. Also, let $d_i = 2i-(n+1)$. For specific $x_p$, notice that if $i > p$, then $|x_i-x_p|$ will result in a negative $x_p$ term. If $i < p$, then $|x_p-x_i|$ will result in a positive $x_p$ term, and otherwise nothing. We now see that the coefficient of $x_p$ will be $$\left((p-1)-1+1\right)-\left(n-(p+1)+1)\right) = 2p-n-1$$as desired. However, $\frac{nd_i}{2} = pn-\frac{n^2}{2}-\frac{n}{2} = \binom{n}{2}-n(n-p) = xc_i$. This means $L = 4\left(\frac{2}{n}\right)^2I^2 = \frac{16}{n^2}I^2$, which proves the inequality. Finally, we see that equality holds when $\frac{|x_i-x_j|}{|i-j|}$ is a constant which gives equality in the Cauchy-Schwarz. Since $i < j \implies x_i < x_j$, this means $x_1, x_2, \ldots x_n$ must be an arithmetic sequence. $\blacksquare$

Remarks: If this problem did not tell you the equality condition, it would be quite harder because you have to guess it, which makes the Cauchy-Schwarz part much more contrived. I actually enjoyed bashing this problem unlike some other problems I did.
This post has been edited 2 times. Last edited by jeteagle, Jul 26, 2021, 3:45 AM
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
TheUltimate123
1740 posts
#19 • 2 Y
Y by Ru83n05, radian_51
Shift so that \(\sum_{i=1}^nx_i=0\). By Cauchy-Schwarz, \[ \left[\sum_{i=1}^n\left(i-\frac{n+1}2\right)x_i\right]^2 \le\frac{n(n+1)(n-1)}{12}\cdot\sum_{i=1}^nx_i^2;\]that is, \[\left[4\sum_{i=1}^n\left(i-\frac{n+1}2\right)x_i\right]^2 \le\frac{2(n^2-1)}3\left[2n\sum_{i=1}^nx_i^2-2\left(\sum_{i=1}^nx_i\right)^2\right],\]which is equivalent to the desired inequality. Equality holds iff \(x_1\), \ldots, \(x_n\) is an arithmetic sequence.
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
awesomeming327.
1696 posts
#20 • 1 Y
Y by solasky
Since shifting the sequence $(x_1,x_2,\dots, x_n)$ does not affect the problem, we can assume that the sum is $0$. We have
\begin{align*}\left(\sum_{i,j=1}^{n}|x_i-x_j|\right)^2 &=\left(2\sum_{1\le i\le j\le n}(x_j-x_i)\right)^2 \\
&= \left((2n-2)x_n+(2n-6)x_{n-1}+\dots +(2-2n)x_1\right)^2 \\
&\le ((2n-2)^2+(2n-6)^2+(2n-10)^2+\dots + (2-2n)^2)(x_1^2+x_2^2+\dots + x_n^2) \\
&= \frac{4(n-1)(n)(n+1)}{3}(x_1^2+x_2^2+\dots + x_n^2) \\
&= \frac{2(n^2-1)}{3}\cdot 2(nx_1^2 + nx_2^2 + \dots + nx_n^2) \\
&= \frac{2(n^2-1)}{3}\cdot 2\left((n-1)\left(\sum_{i=1}^{n}{x_i^2}\right) + \left(\sum_{i=1}^{n}{x_i}\right)^2 - 2\sum_{1\le i<j\le n}x_ix_j\right) \\
&= \frac{2(n^2-1)}{3}\cdot 2\left(\sum_{1\le i<j\le n}(x_i-x_j)^2\right) \\
&= \frac{2(n^2-1)}{3}\sum_{i,j=1}^{n}(x_i-x_j)^2
\end{align*}Clearly, the equality condition of Cauchy will force the equality case of this inequality to also be an arithmetic sequence. In the case that one of the coefficients in the second line of our work is zero, we can add $\varepsilon$ to each of the coefficients to get the same inequality but avoid division by zero.
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
solasky
1566 posts
#21 • 1 Y
Y by radian_51
Induction led me to a solution that is way too long, but I’d recommend you to try it if you couldn’t find the Cauchy-Schwarz way (like me)

We will do induction on $n$. The base case $n = 1$ simplifies to $0 \le 0$, which is clear. For the inductive step, we will assume that the problem is true for $n = k$, and prove that it is also true for $n = k + 1$. Noticing that the problem statement does not change if we shift all variables $x_i$ by a constant, assume that their sum $\sum_{i = 1}^k x_i = 0$ to simplify calculations. Let \[\rho_1 = \sum_{i, j=1}^k |x_i - x_j| \text{ and } \rho_2 = \sum_{i,j = 1}^k (x_i - x_j)^2.\]By the inductive hypothesis, we may assume that \[\frac{2(k^2 - 1)}{3} \cdot \rho_2 \ge \rho_1^2.\]Let $P(x_{k + 1})$ be \[P(x_{k + 1}) = \frac{2[(k + 1)^2 - 1]}{3} \cdot \sum_{i,j = 1}^{k + 1} (x_i - x_j)^2 - \left(\sum_{i,j = 1}^{k + 1}|x_i - x_j|\right)^2.\]We wish to show that $P(x_{k + 1}) \ge 0$ for $x_{k + 1} \ge x_k$. Using the fact that $\sum_{i = 1}^k x_i = 0$, we can calculate \[\sum_{i,j = 1}^{k + 1} |x_i - x_j| = p_1 + 2 \sum_{i = 1}^{k} (x_{k + 1} - x_i) = p_1 + 2kx_{k + 1}.\]We can also calculate \[\sum_{i,j = 1}^{k + 1}(x_i - x_j)^2 = \rho_2 + 2\sum_{i = 1}^k (x_i - x_{k + 1})^2 = p_2 + 2\sum_{i = 1}^k x_i^2 + 2k x_{k + 1}^2.\]In addition, I claim that \[2\sum_{i = 1}^k x_i^2 = \frac{\rho_2}{k}.\]This is because \begin{align*}
        \rho_2 &= \sum_{i,j = 1}^k (x_i - x_j)^2 \\
        &= 2k \sum_{i = 1}^k x_i^2 - 2\sum_{i,j = 1}^k x_i x_j \\
        &= 2k \sum_{i = 1}^k x_i^2 - 2 \left(\sum_{i = 1}^k x_i\right)\left(\sum_{j = 1}^k x_j\right) \\
        &= 2k \sum_{i = 1}^k x_i^2.
    \end{align*}So, we can further simplify our formula to get \[\sum_{i,j = 1}^{k + 1} (x_i - x_j)^2 = \frac{k + 1}{k} \cdot \rho_2 + 2kx_{k + 1}^2.\]Thus, \[P(x_{k + 1}) = \frac{2k^2 + 4k}{3} \cdot \left(\frac{k + 1}{k} \cdot \rho_2 + 2kx_{k + 1}^2\right) - (\rho_1 + 2k x_{k + 1})^2.\]Treating this as a quadratic polynomial in terms of $x_{k + 1}$ holding all other terms are constant, $P(x_{k + 1})$ is minimized at \[x_{k + 1} = \frac{-(-4k\rho_1)}{2\left(\frac{4k^3 + 8k^2}{3} - 4k^2\right)} = \frac{4k\rho_1}{2\left(\frac{4k^3 - 4k^2}{3}\right)} = \frac{3\rho_1}{2k(k - 1)}.\]Plugging this in, \[P\left(\frac{3\rho_1}{2k(k - 1)}\right) = \frac{2k^2 + 4k}{3} \cdot \left(\frac{k + 1}{k} \cdot \rho_2 + \frac{9\rho_1^2}{2k(k - 1)^2}\right) - \left(\rho_1 + \frac{3\rho_1}{k - 1}\right)^2.\]This simplifies to \begin{align*}
        P(x_{k + 1}) &= \frac{2(k + 2)(k + 1)}{3} \cdot \rho_2 + \left(\frac{3(k + 2)}{(k - 1)^2} - \frac{(k + 2)^2}{(k - 1)^2}\right)\rho_1^2 \\
        &= \frac{k + 2}{k - 1}\left(\frac{2(k^2 - 1)}{3}\rho_2 - \rho_1^2\right).
    \end{align*}By the inductive hypothesis, $\frac{2(k^2 - 1)}{3}\rho_2 \ge \rho_1^2$, so we get that $P(x_{k + 1}) \ge 0$ which completes the inductive step. Equality holds if and only if for all $0 \le k \le n - 1$, \[x_{k + 1} = \frac{3\sum_{i,j = 1}^k |x_i - x_j|}{2k(k - 1)}.\]We will prove that this is equivalent to $x_1, x_2, \ldots, x_n$ being an arithmetic sequence using induction. For $n = 2$, the problem statement becomes $4(x_2 - x_1)^2 \le 4(x_2 - x_1)^2$, which is true for all $x_1$ and $x_2$. Since any two numbers are in an arithmetic sequence, the base case is clear. For the inductive step, assume that $d = x_2 - x_1 = x_3 - x_2 = \cdots = x_k - x_{k - 1}$, and we will prove that $x_{k + 1} - x_k = d$ as well. This is equivalent to the numbers being in an arithmetic sequence. Since we previously assumed that $x_1 + x_2 + \cdots + x_k = 0$, this means that $x_k = \frac{(k - 1)d}{2}$. We can simplify \begin{align*}
        x_{k + 1} &= \frac{3 \sum_{i,j = 1}^k |x_i - x_j|}{2k(k - 1)} \\
        &= \frac{3 \sum_{1 \le i < j \le k} d(j - i)}{k(k - 1)}.
        \end{align*}In the sum in the numerator, a difference of $d$ occurs $n - 1$ times, a difference of $2d$ occurs $n - 2$ times, and so-on. Thus, \begin{align*}
        x_{k + 1} &= \frac{3d}{(k - 1)k}((k - 1) \cdot 1 + (k - 2) \cdot 2 + \cdots + 1 \cdot (k - 1)) \\
        &= \frac{3d}{(k - 1)k} \sum_{i = 1}^{k - 1} i(k - i) \\
        &= \frac{3d}{(k - 1)k} \left(k\sum_{i = 1}^{k - 1} i - \sum_{i = 1}^{k - 1} i^2 \right) \\
        &= \frac{3d}{(k - 1)k} \left(k \cdot \frac{(k - 1)k}{2} - \frac{(k - 1)k(2k - 1}{6}\right) \\
        &= 3d \left(\frac{k}{2} - \frac{2k - 1}{6}\right) \\
        &= d\left(\frac{k + 1}{2}\right).
    \end{align*}Thus, $x_{k + 1} - x_k = d$, completing the inductive step. Thus, we have proven the inequality and have shown that equality holds if and only if the variables are in an arithmetic sequence.
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
sami1618
892 posts
#22
Y by
Divide both sides by $4$ and only consider the case when $i<j$. Let $a_i=x_{i+1}-x_i$. Then the second summation becomes, $$\sum_{i=1}^{n-1}\sum_{j=i}^{n-1}(a_i+a_{i+1}+\dots+a_j)^2$$Since equality occurs when $a_1=a_2=\dots=a_{n-1}$ we apply the Cauchy Schwartz inequality in the form $$(\frac{n}{2}\sum_{i=1}^{n-1}\sum_{j=i}^{n-1}(a_i+a_{i+1}+\dots+a_j))^2\leq(\sum_{i=1}^{n-1}\sum_{j=i}^{n-1}(a_i+a_{i+1}+\dots+a_j)^2)(\sum_{i=1}^{n-1}\sum_{j=i}^{n-1}(i-j)^2)$$$$(\sum_{i=1}^{n-1}\sum_{j=i}^{n-1}(a_i+a_{i+1}+\dots+a_j))^2\leq \frac{n^2-1}{3}\sum_{i=1}^{n-1}\sum_{j=i}^{n-1}(a_i+a_{i+1}+\dots+a_j)^2$$
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
pie854
243 posts
#23 • 1 Y
Y by radian_51
Let's set $x_i=a_{i-1}+\dots+a_1+x_1$, where the $a_j$'s are non-negative. Then we have \begin{align*} \frac 23 (n^2-1) \sum_{i=1}^n \sum_{j=1}^n (x_i-x_j)^2 & = \frac 43 (n^2-1) \sum_{i=2}^n \sum_{j=1}^{i-1} (x_i-x_j)^2 \\ & = \frac 43 (n^2-1) \sum_{i=2}^n \sum_{j=1}^{i-1} (a_{i-1}+a_{i-2}+\dots+a_j)^2 \\ & = \frac 43 (n^2-1) \left [ \sum_{k=1}^{n-1} k(n-k) a_k^2+2\sum_{i=1}^{n-2}\sum_{j=i+1}^{n-1} i(n-j) a_i a_j \right ].\end{align*}And similarly, \begin{align*} \left (\sum_{i=1}^n \sum_{j=1}^n |x_i-x_j|\right)^2 & = 4 \left (\sum_{i=2}^n \sum_{j=1}^{i-1} (x_i-x_j)\right)^2 \\ & =  4 \left (\sum_{i=2}^n \sum_{j=1}^{i-1} (a_{i-1}+a_{i-2}+\dots+a_j)\right)^2 \\ & = 4\left [ \sum_{k=1}^{n-1} k^2(n-k)^2 a_k^2+2\sum_{i=1}^{n-2}\sum_{j=i+1}^{n-1} ij(n-i)(n-j) a_i a_j \right].\end{align*}original fakesolve; kept for amusement ;)

So we need to prove that $$F(a_1,a_2,\dots,a_{n-1}):=\sum_{k=1}^{n-1} \left [ k(n-k)(n^2-1)-3k^2(n-k)^2\right ]a_k^2 -\sum_{i=1}^{n-2}\sum_{j=i+1}^{n-1} \left [ 3ij(n-i)(n-j)-i(n^2-1)(n-j)\right ] 2a_ia_j\geqslant 0 \qquad (1)$$The inequality is homogeneous, so we can assume wlog that $$a_1+\dots+a_{n-1}=1\qquad (2)$$So that the domain of $F$ becomes $[0,1]^{n-1}$ which is compact. So $F$ attains a global minimum.

First, suppose $\{a_1,\dots,a_{n-1}\}\subset \{0,1\}$. If one of the $a_i$'s is $1$ and the rest are $0$, then $F>0$ because $k(n-k)(n^2-1)-3k^2(n-k)^2>0$ is true when $1\leq k<n$. And when all are $0$ then obviously $F=0$. Now suppose $a_1,\dots,a_{n-1}\in (0,1)$ so that we can proceed with Lagrange Multipliers. Taking the gradients of $F$ and the constraint function $a_1+\dots+a_{n-1}$ we get the equations \begin{align*} & 2(k(n-k)(n^2-1)-3k^2(n-k)^2)a_k-2\sum_{i=1}^{k-1} (3ik(n-i)(n-k)-i(n^2-1)(n-k))a_i \\ & \qquad -2\sum_{j=k+1}^{n-1} (3kj(n-k)(n-j)-k(n^2-1)(n-j))a_j=\lambda, \quad (k=1,2,\dots, n-1) \qquad (3)\end{align*}(Take the empty sum to be $0$). Notice that we have $n$ variables $\lambda,a_1,\dots,a_{n-1}$ and $n$ linear independent (this is not very hard to see?) equations $(2)$ and $(3)$. So if there is a solution, it must be unique. We claim the solution is $$a_1=a_2=\dots=a_{n-1}=\frac{1}{n-1}, \quad \lambda=0.$$Indeed this solution satisfies $(2)$ and we can show without much effort that $$\sum_{i=1}^{k-1} (3ik(n-i)(n-k)-i(n^2-1)(n-k))+\sum_{j=k+1}^{n-1} (3kj(n-k)(n-j)-k(n^2-1)(n-j))=k(n-k)(n^2-1)-3k^2(n-k)^2 \qquad (4)$$so it also satisfies $(3)$. Finally, showing $F\left (\frac{1}{n-1},\frac{1}{n-1},\dots,\frac{1}{n-1}\right)=0$ is pretty much the same as proving $(4)$, so we'll omit this too.

Thus the global minimum of $F$ is $0$. This proves $(1)$ and so the original inequality as well. Also the equality holds iff $a_1=\dots=a_{n-1}$, as seen above. This implies that the equality case of the original inequality is when $x_1, \dots, x_n$ is arithmetic. This finishes the problem.
This post has been edited 8 times. Last edited by pie854, Sep 8, 2024, 10:31 AM
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
Ahmqd
3 posts
#24 • 1 Y
Y by radian_51
pie854 wrote:
First, suppose $\{a_1,\dots,a_{n-1}\}\subset \{0,1\}$. If one of the $a_i$'s is $1$ and the rest are $0$, then $F>0$ because $k(n-k)(n^2-1)-3k^2(n-k)^2>0$ is true when $1\leq k<n$. And when all are $0$ then obviously $F=0$. Now suppose $a_1,\dots,a_{n-1}\in (0,1)$

Don't we also have the case that, say, $a_3=0$ and the other $a_i$ are arbitrary with sum $1$?
Z K Y
The post below has been deleted. Click to close.
This post has been deleted. Click here to see post.
eevee9406
5 posts
#25
Y by
But for $a_i=2i, n=4$ the equality doesn't hold?

Also (and I might be wrong) moving the RHS sum to LHS allows for Titu's Lemma by removing terms equal to zero, quicksolve?
This post has been edited 1 time. Last edited by eevee9406, Apr 15, 2025, 4:50 AM
Z K Y
N Quick Reply
G
H
=
a